Low-density parity-check codes for asymmetric distributed source coding

Abstract

Low-Density Parity-Check (LDPC) codes achieve good performance, tending towards the Slepian-Wolf bound, when used as channel codes in Distributed Source Coding (DSC). Most LDPC codes found in literature are designed assuming random distribution of transmission errors. However, certain DSC applications can predict the error location within a certain level of accuracy. This feature can be exploited in order to design application specific LDPC codes to enhance the performance of traditional LDPC codes. This paper proposes a novel architecture for asymmetric DSC where the encoder is able to estimate the location of the errors within the side information. It then interleaves the bits having a high probability of error to the beginning of the codeword. The LDPC codes are designed to provide a higher level of protection to the front bits. Simulation results show that correct localization of errors pushes the performance of the system on average 13.3% closer to the Slepian-Wolf bound, compared to the randomly constructed LDPC codes. If the error localization prediction fails, such that the errors are randomly distributed, the performance is still in line with that of the traditional DSC architecture.